How production un-ready are Mesos Cassandra, Spark and Kafka Frameworks?

[Chris Elsmore <ch...@demandlogic.co.uk>]

Hi all,

Have just got back from a brilliant MesosCon Europe in Dublin, I learnt a huge amount and a big thank-you for putting on a great conference to all involved!


I am looking to deploy a small (maybe 5 max) Cassandra & Spark cluster to do some data analysis at my current employer, and am a little unsure of the current status of the frameworks this would need to run on Mesos- both the mesosphere docs (which I’m guessing use the frameworks of the same name hosted on Github) and the Github ReadMes mention that these are not production ready, and the rough timeline of Q1 2016.

I’m just wondering how production un-ready these are!? I am looking at using Mesos to deploy future stateless services in the next 6 months or so, and so I like the idea of adding to that system and the look of the configuration that is handled for you to bind nodes together in these frameworks. However it feels like for a smallish cluster of production ready machines it might be better to deploy them standalone and stay observant on the status of such things in the near future, and the configuration wins are not that large especially for a small cluster.


Any experience and advice on the above would be greatly received!


Chris

Re: How production un-ready are Mesos Cassandra, Spark and Kafka Frameworks?

[Sam Bessalah <sa...@gmail.com>]

Thats' really depends on what you're doing.
I've been running Spark in production with Mesos for as far as Spark ever
got open sourced.
Earlier this year, we added Cassandra in the mix by running it through
Docker and Marathon in host network mode with volume. Nothing fancy, since
it was for non critical user facing application. I have been experimenting
with the Cassandra mesos framework for sometime, it still has some rough
edges, but with our analytics operations, I will be pushing into prod very
soon.
Running Cassandra in containers is a rather complicated task, because of
the number of configurations Cassandra adds to the linux kernel. But
performance wise I haven't seen any real big difference. My main issue has
been with constraints to mange to maintain some data locality with Spark
and C*. Spark has lately added the support for Mesos Constraints, but I
have yet to experiment with that.
As with Kafka, It's been running without any issue so far. But I want to
see how it will handle persistent primitives in Mesos.



On Tue, Oct 13, 2015 at 3:34 PM, Chris Elsmore <
chris.elsmore@demandlogic.co.uk> wrote:

> Thanks for the replies, a few other people IRL have echoed Dicks comments.
>
> To take advantage of Cassandra and Spark data locality, I guess I could
> run Cassandra on a few Mesos agent machines outside of Mesos, and just
> label them such that spark jobs get assigned to those too via Mesos?
>
>
>
>
> On 13 Oct 2015, at 08:38, craig w <co...@gmail.com> wrote:
>
> So far the Kafka framework has worked well in production. We launch the
> framework using marathon, then execute a few command line statements to add
> and start the brokers.
> On Oct 12, 2015 1:38 PM, "Dick Davies" <di...@hellooperator.net> wrote:
>
>> Hi Chris
>>
>> <personal opinion ahoy>
>>
>> Spark is a Mesos native, I'd have no hesitation running it on Mesos.
>>
>> Cassandra not so much -
>> that's not to disparage the work people are putting in there, I think
>> it's really interesting. But personally with complex beasts like Cassandra
>> I want to be running as 'stock' as possible, as it makes it easier to
>> learn
>> from other peoples experiences.
>>
>> On 12 October 2015 at 17:47, Chris Elsmore <
>> chris.elsmore@demandlogic.co.uk> wrote:
>>
>>> Hi all,
>>>
>>> Have just got back from a brilliant MesosCon Europe in Dublin, I learnt
>>> a huge amount and a big thank-you for putting on a great conference to all
>>> involved!
>>>
>>>
>>> I am looking to deploy a small (maybe 5 max) Cassandra & Spark cluster
>>> to do some data analysis at my current employer, and am a little unsure of
>>> the current status of the frameworks this would need to run on Mesos- both
>>> the mesosphere docs (which I’m guessing use the frameworks of the same name
>>> hosted on Github) and the Github ReadMes mention that these are not
>>> production ready, and the rough timeline of Q1 2016.
>>>
>>> I’m just wondering how production un-ready these are!? I am looking at
>>> using Mesos to deploy future stateless services in the next 6 months or so,
>>> and so I like the idea of adding to that system and the look of the
>>> configuration that is handled for you to bind nodes together in these
>>> frameworks. However it feels like for a smallish cluster of production
>>> ready machines it might be better to deploy them standalone and stay
>>> observant on the status of such things in the near future, and the
>>> configuration wins are not that large especially for a small cluster.
>>>
>>>
>>> Any experience and advice on the above would be greatly received!
>>>
>>>
>>> Chris
>>>
>>>
>>>
>>>
>>
>

Re: How production un-ready are Mesos Cassandra, Spark and Kafka Frameworks?

[Chris Elsmore <ch...@demandlogic.co.uk>]

Thanks for the replies, a few other people IRL have echoed Dicks comments.

To take advantage of Cassandra and Spark data locality, I guess I could run Cassandra on a few Mesos agent machines outside of Mesos, and just label them such that spark jobs get assigned to those too via Mesos?




> On 13 Oct 2015, at 08:38, craig w <co...@gmail.com> wrote:
> 
> So far the Kafka framework has worked well in production. We launch the framework using marathon, then execute a few command line statements to add and start the brokers.
> 
> On Oct 12, 2015 1:38 PM, "Dick Davies" <dick@hellooperator.net <ma...@hellooperator.net>> wrote:
> Hi Chris
> 
> <personal opinion ahoy>
> 
> Spark is a Mesos native, I'd have no hesitation running it on Mesos.
> 
> Cassandra not so much -
> that's not to disparage the work people are putting in there, I think
> it's really interesting. But personally with complex beasts like Cassandra
> I want to be running as 'stock' as possible, as it makes it easier to learn
> from other peoples experiences.
> 
> On 12 October 2015 at 17:47, Chris Elsmore <chris.elsmore@demandlogic.co.uk <ma...@demandlogic.co.uk>> wrote:
> Hi all,
> 
> Have just got back from a brilliant MesosCon Europe in Dublin, I learnt a huge amount and a big thank-you for putting on a great conference to all involved!
> 
> 
> I am looking to deploy a small (maybe 5 max) Cassandra & Spark cluster to do some data analysis at my current employer, and am a little unsure of the current status of the frameworks this would need to run on Mesos- both the mesosphere docs (which I’m guessing use the frameworks of the same name hosted on Github) and the Github ReadMes mention that these are not production ready, and the rough timeline of Q1 2016.
> 
> I’m just wondering how production un-ready these are!? I am looking at using Mesos to deploy future stateless services in the next 6 months or so, and so I like the idea of adding to that system and the look of the configuration that is handled for you to bind nodes together in these frameworks. However it feels like for a smallish cluster of production ready machines it might be better to deploy them standalone and stay observant on the status of such things in the near future, and the configuration wins are not that large especially for a small cluster.
> 
> 
> Any experience and advice on the above would be greatly received!
> 
> 
> Chris
> 
> 
> 
> 

Re: How production un-ready are Mesos Cassandra, Spark and Kafka Frameworks?

[craig w <co...@gmail.com>]

So far the Kafka framework has worked well in production. We launch the
framework using marathon, then execute a few command line statements to add
and start the brokers.
On Oct 12, 2015 1:38 PM, "Dick Davies" <di...@hellooperator.net> wrote:

> Hi Chris
>
> <personal opinion ahoy>
>
> Spark is a Mesos native, I'd have no hesitation running it on Mesos.
>
> Cassandra not so much -
> that's not to disparage the work people are putting in there, I think
> it's really interesting. But personally with complex beasts like Cassandra
> I want to be running as 'stock' as possible, as it makes it easier to learn
> from other peoples experiences.
>
> On 12 October 2015 at 17:47, Chris Elsmore <
> chris.elsmore@demandlogic.co.uk> wrote:
>
>> Hi all,
>>
>> Have just got back from a brilliant MesosCon Europe in Dublin, I learnt a
>> huge amount and a big thank-you for putting on a great conference to all
>> involved!
>>
>>
>> I am looking to deploy a small (maybe 5 max) Cassandra & Spark cluster to
>> do some data analysis at my current employer, and am a little unsure of the
>> current status of the frameworks this would need to run on Mesos- both the
>> mesosphere docs (which I’m guessing use the frameworks of the same name
>> hosted on Github) and the Github ReadMes mention that these are not
>> production ready, and the rough timeline of Q1 2016.
>>
>> I’m just wondering how production un-ready these are!? I am looking at
>> using Mesos to deploy future stateless services in the next 6 months or so,
>> and so I like the idea of adding to that system and the look of the
>> configuration that is handled for you to bind nodes together in these
>> frameworks. However it feels like for a smallish cluster of production
>> ready machines it might be better to deploy them standalone and stay
>> observant on the status of such things in the near future, and the
>> configuration wins are not that large especially for a small cluster.
>>
>>
>> Any experience and advice on the above would be greatly received!
>>
>>
>> Chris
>>
>>
>>
>>
>

Re: How production un-ready are Mesos Cassandra, Spark and Kafka Frameworks?

[Dick Davies <di...@hellooperator.net>]

Hi Chris

<personal opinion ahoy>

Spark is a Mesos native, I'd have no hesitation running it on Mesos.

Cassandra not so much -
that's not to disparage the work people are putting in there, I think
it's really interesting. But personally with complex beasts like Cassandra
I want to be running as 'stock' as possible, as it makes it easier to learn
from other peoples experiences.

On 12 October 2015 at 17:47, Chris Elsmore <ch...@demandlogic.co.uk>
wrote:

> Hi all,
>
> Have just got back from a brilliant MesosCon Europe in Dublin, I learnt a
> huge amount and a big thank-you for putting on a great conference to all
> involved!
>
>
> I am looking to deploy a small (maybe 5 max) Cassandra & Spark cluster to
> do some data analysis at my current employer, and am a little unsure of the
> current status of the frameworks this would need to run on Mesos- both the
> mesosphere docs (which I’m guessing use the frameworks of the same name
> hosted on Github) and the Github ReadMes mention that these are not
> production ready, and the rough timeline of Q1 2016.
>
> I’m just wondering how production un-ready these are!? I am looking at
> using Mesos to deploy future stateless services in the next 6 months or so,
> and so I like the idea of adding to that system and the look of the
> configuration that is handled for you to bind nodes together in these
> frameworks. However it feels like for a smallish cluster of production
> ready machines it might be better to deploy them standalone and stay
> observant on the status of such things in the near future, and the
> configuration wins are not that large especially for a small cluster.
>
>
> Any experience and advice on the above would be greatly received!
>
>
> Chris
>
>
>
>